On using Queueing Network Models with finite capacity queues for Software Architectures performance prediction

We are interested in studying the performance of software system in early stages of development. We investigate how queueing networks with finite capacity and blocking can be applied as performance models of software architectures. The starting point of our analysis is a software system high-level description whose dynamic behavior results in a finite state model or, as recently proposed, is described in terms of Message Sequence Charts. We use the queuing theory to obtain a performance model of the software component behaviour from the behavioral description. Our approach is to automatically derive a queuing network from a Software Architecture description. More precisely, we represent a software component or a set of components with a simple server or complex server. A complex server represents the service given by the associated components, where the associated service time is the summation of the single service time associated to each component. For particular behavioral patterns we use also multiclass server that represents a server that can provide multiple service but only one at a time. In the examples we considered this class of queuing network models was not sufficiently expressive. In fact by using only service centers with infinite capacities we cannot model systems where there are concurrent components that can communicate synchronously. Hence, by observing the need of a more accurate definition of the performance models of software architectures to capture some features of the communication systems, we consider queueing networks with finite capacity and blocking to represent some synchronization constraints. To model synchronous communication among concurrent system components, we assigns distinct service centers to the communicating components in order to model their independence. We associate to the receiver component a service center with a zero capacity buffer and impose a blocking mechanism to the sender component in order to model synchronization. In particular we choose the Blocking After Service (BAS) blocking mechanism. Thanks to this kind of modeling we can describe more complex contexts where the components are simultaneously active but also situations in which a component C1 attempts to communicate with another component C2, when the latter is still working. Indeed, the BAS mechanism permits to block component C1 waiting for C2 to complete its service. In the paper we will discuss our experience in using queueing network models and their adequacy for the problem at hand in terms both of modeling and of its evaluation.